Find
Facts are reachable, current, consistent and available in an agent-usable form.
Methodology
A transparent AI-agent testing methodology.
Versioned scenarios, deterministic assertions and reviewable evidence keep each result accountable.
Measure the customer journey, not the presence of a tag.
A readiness signal combines discoverability, successful task completion, repeatability and efficiency.
Understand
Controls, policies, prices and states have unambiguous meaning.
Act
The allowed workflow reaches an independently verified final state.
Recover
Validation, auth, session and inventory failures expose a workable next action.
Coverage is separated by channel and access mode.
A missing agent-native surface cannot erase a successful browser result, and the reverse is also true.
AI search
Crawl access, citation surfaces, content consistency and machine-readable context.
Browser and UI
Rendered interaction, semantic controls, keyboard path, sessions and responsive behavior.
API and MCP
Schemas, authorization, errors, idempotency, permissions and confirmation boundaries.
Public
Unauthenticated surfaces with irreversible actions excluded.
Authenticated
Approved test accounts, fixtures and redacted evidence.
Sandbox
Complete journeys using controlled data and reversible operations.
The scenario is a versioned test contract.
Prompt wording alone never defines success. The contract states the goal, permissions, stop gate and assertions.
- Persona and business goal
- Prerequisites and test data
- Allowed actions and production stop gate
- Independent final-state assertions
- Required evidence and criticality
Record the path and the cost of completion.
Session evidence explains whether a nominal success was fast, stable, repeatable and autonomous.
- Video or event-aligned session replay
- Step trace, screenshots, DOM and accessibility tree
- Total duration, step latency and waiting time
- Retries, recovery attempts and human interventions
- Page CLS and movement of the intended target
CLS is recorded with interaction context.
Page-level Cumulative Layout Shift is useful, but target stability is also inspected around each planned click or input.
Repeatability protocol
Standard tasks run three times. Critical tasks run five times. Agent, model, tools, viewport, locale, auth state and policy restrictions remain attached to every run.
Separate website defects from executor noise.
Results distinguish website, agent-specific, policy and infrastructure failure before scoring.
passpartialfailblocked
Readiness and confidence are different numbers.
Readiness reflects observed capability. Confidence reflects coverage, repetitions, evidence quality and unresolved uncertainty.
P0 blocks a critical task or creates material safety risk.
P1 creates major repeatable friction or unreliable completion.
Evidence is useful only when it is safe to review.
Artifacts are stored with stable IDs, timestamps, environment and retention rules.
- Screenshots, recording, accessibility tree, DOM, console, network and API traces
- Redaction for PII, credentials, tokens, payment, financial and medical data
- Purpose-limited retention and controlled access
Emerging technologies are inventory signals, not automatic grades.
WebMCP, MCP, OpenAPI, llms.txt, llms-full.txt and other agent surfaces are detected and lifecycle-labeled. Their absence is not automatically a failure.
Current limitations
Agent products, policies and models change. Results are versioned observations from a defined environment, not permanent compatibility guarantees.
Use a method your team can inspect.
Start with a scoped audit and keep every conclusion connected to evidence.